def test_faster_rcnn_mobile_param_groups():
backbone = models.torchvision.faster_rcnn.backbones.mobilenet(
pretrained=False)
model = faster_rcnn.model(num_classes=6, backbone=backbone)
param_groups = model.param_groups()
assert len(param_groups) == 6
def test_e2e_detect(samples_source, fridge_class_map, model_name,
param_groups_len):
img_path = samples_source / "fridge/odFridgeObjects/images/10.jpg"
tfms_ = tfms.A.Adapter([A.Resize(384, 384), A.Normalize()])
backbone_fn = getattr(models.torchvision.faster_rcnn.backbones, model_name)
backbone = backbone_fn(pretrained=False)
model = faster_rcnn.model(num_classes=4, backbone=backbone)
pred_dict = faster_rcnn.end2end_detect(img_path,
tfms_,
model,
fridge_class_map,
detection_threshold=1)
assert len(pred_dict["detection"]["bboxes"]) == 0
def fridge_faster_rcnn_model() -> nn.Module:
backbone = faster_rcnn.backbones.resnet_fpn.resnet18(pretrained=False)
return faster_rcnn.model(num_classes=5, backbone=backbone)
def test_faster_rcnn_default_param_groups():
model = faster_rcnn.model(num_classes=4)
param_groups = model.param_groups()
assert len(param_groups) == 8
def fridge_faster_rcnn_model() -> nn.Module:
backbone = models.torchvision.faster_rcnn.backbones.resnet18_fpn(pretrained=False)
return faster_rcnn.model(num_classes=5, backbone=backbone)
def test_faster_rcnn_fpn_backbones(model_name, param_groups_len):
backbone_fn = getattr(models.torchvision.faster_rcnn.backbones, model_name)
backbone = backbone_fn(pretrained=False)
model = faster_rcnn.model(num_classes=4, backbone=backbone)
assert len(model.param_groups()) == param_groups_len